Alleviating the Sparse Matrix Scaling Bottleneck in Adaptive VQE via High-Order Taylor State Evolution

Hermawan Kresno Dipojono

arxiv: 2606.29692 · v1 · pith:QXFGGMOMnew · submitted 2026-06-29 · 🪐 quant-ph

Alleviating the Sparse Matrix Scaling Bottleneck in Adaptive VQE via High-Order Taylor State Evolution

Hermawan Kresno Dipojono This is my paper

Pith reviewed 2026-06-30 06:35 UTC · model grok-4.3

classification 🪐 quant-ph

keywords adaptive VQETaylor series expansionsparse matrix-vector multiplicationvariational quantum eigensolverquantum chemistry simulationNISQ algorithmsfermion-to-qubit mapping

0 comments

The pith

A fifth-order Taylor series replaces dense matrix exponentiation in adaptive VQE, scaling state updates linearly with non-zero elements while preserving subchemical accuracy.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper demonstrates that the classical bottleneck in adaptive VQE, caused by building and exponentiating ever-larger sparse operator matrices during ansatz growth, can be removed by evaluating each state update through five successive sparse matrix-vector multiplications from a Taylor series. This keeps computational cost proportional only to the number of non-zero matrix entries and still delivers state fidelities above 0.999999 together with ground-state energies accurate to subchemical precision on BeH2 and H2O molecules under both Jordan-Wigner and Bravyi-Kitaev mappings. The approach is validated on systems whose matrices contain more than 268 million structural elements. A reader would care because the method removes the memory-bandwidth and CPU-cycle limits that currently cap how deep an adaptive ansatz can be grown on classical simulators.

Core claim

By reframing unitary evolution as a deterministic fifth-order Taylor expansion performed through five chained sparse matrix-vector multiplications, the method achieves O(Nz) scaling for state updates, maintains state fidelity greater than 0.999999, and recovers absolute ground-state energies to subchemical accuracy on equilibrium BeH2 (14 qubits), equilibrium H2O (12 qubits), and stretched H2O profiles.

What carries the argument

Fifth-order Taylor series expansion for the state update under unitary evolution, evaluated as five successive sparse matrix-vector multiplications.

If this is right

Ansatz growth in adaptive VQE can continue with far lower classical memory and compute demands during the simulation phase.
Deeper variational circuits become simulable on systems whose operator matrices exceed hundreds of millions of elements.
The same update routine works unchanged for both Jordan-Wigner and Bravyi-Kitaev mappings while retaining subchemical accuracy on equilibrium and stretched geometries.
The framework supports high-fidelity variational simulations on hardware platforms with constrained computational budgets.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The linear-scaling update could be applied to other variational algorithms that repeatedly apply operators during optimization loops.
For still larger molecules or deeper ansatzes the fixed fifth-order truncation might eventually need to be replaced by an adaptive order that monitors local error.
Standard laptop or small-cluster hardware could now run adaptive VQE instances that previously required specialized high-memory servers for matrix handling.

Load-bearing premise

The truncation error from stopping the Taylor series at fifth order does not accumulate across successive operator additions in the adaptive loop or shift the minimized energy outside chemical accuracy for the tested molecules and mappings.

What would settle it

Running the identical adaptive VQE procedure with exact matrix exponentiation or a higher-order Taylor expansion on the same BeH2 and H2O instances and checking whether the final energies differ by more than chemical accuracy or the fidelities drop below 0.999999.

read the original abstract

The Variational Quantum Eigensolver (VQE) is a leading algorithm for noisy intermediate-scale quantum (NISQ) devices, but its adaptive variants (e.g., ADAPT-VQE) suffer from severe classical simulation bottlenecks during the ansatz growth phase. Representing and exponentiating pool operators for multi-qubit systems constructs massive sparse matrices that quickly scale to millions of elements, choking classical memory bandwidth and CPU/GPU cycle capacity. In this work, we present a resource-efficient software-layer framework that completely bypasses dense matrix exponentiation by evaluating state updates through a deterministic fifth-order (O(5)) Taylor series expansion. This approach reframes the costly unitary evolution into a chained sequence of five lightweight, successive sparse matrix-vector multiplications scaling strictly as O(Nz), where Nz is the number of non-zero elements. We validate our framework using equilibrium BeH2 (14 qubits), equilibrium H2O (12 qubits), and strongly correlated asymmetrically stretched H2O molecular profiles under both Jordan-Wigner (JW) and Bravyi-Kitaev (BK) transformations. The simulation results demonstrate that the O(5) truncation maintains exceptional numerical fidelity, retaining a state fidelity > 0.999999 and matching absolute ground state energies down to subchemical accuracy, while effortlessly navigating matrix spaces exceeding 268 million structural elements. This framework provides a scalable, high-performance pathway for executing deep variational simulations on hardware platforms with constrained computational budgets.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

A fixed fifth-order Taylor expansion replaces the matrix exponential inside the ADAPT-VQE loop and still delivers sub-chemical accuracy on the three small molecules they ran.

read the letter

The main point is that five successive sparse matrix-vector products can stand in for the full unitary update during ansatz growth without wrecking the final energies or state overlap on BeH2 (14 qubits) and H2O (12 qubits) under both JW and BK mappings. The numbers they give—fidelity above 0.999999 and energies inside chemical accuracy, even in the stretched H2O regime—come from running the complete adaptive selection loop, so the truncation error is already tested in the setting that matters.

The practical gain is clear: the method scales strictly with the number of non-zeros and sidesteps the memory bandwidth hit from building and exponentiating the pool operators. That is a legitimate engineering fix for groups that simulate adaptive VQE classically and hit the sparse-matrix wall at a few hundred million elements.

The evidence is direct but narrow. They report results only for these three molecular profiles and fix the Taylor order at five; there is no sweep over lower or higher orders and no larger active-space test where the truncation might start to matter. The abstract also gives no error bars or explicit description of how the fidelity was computed against the reference. Those gaps are real but secondary to the central claim, which the data already address.

This is useful reading for anyone who needs to push classical emulations of ADAPT-VQE a bit further on modest hardware. The work is straightforward, the numerical checks are concrete, and the method is reproducible from the description. It should go to peer review.

Referee Report

1 major / 2 minor

Summary. The manuscript proposes replacing dense matrix exponentiation in ADAPT-VQE with a deterministic fifth-order Taylor series expansion for state updates, reducing the cost to five successive sparse matrix-vector multiplications. It reports numerical results on equilibrium BeH2 (14 qubits), equilibrium H2O (12 qubits), and stretched H2O under both Jordan-Wigner and Bravyi-Kitaev mappings, claiming state fidelity above 0.999999 and subchemical accuracy in ground-state energies while handling matrices with over 268 million elements.

Significance. If the numerical results hold, the approach removes a concrete classical bottleneck in adaptive variational algorithms, enabling deeper ansatz growth on classical hardware for systems whose sparse operators exceed current memory-bandwidth limits. The fact that the reported fidelities and energies already incorporate the full adaptive operator-selection loop provides direct evidence against accumulation of truncation error for the tested molecules and mappings.

major comments (1)

[Results] Results section (fidelity and energy tables): the reported state fidelity >0.999999 and subchemical accuracy lack any description of the fidelity computation method, any error-bar analysis, and any comparison against fourth- or sixth-order truncations. These omissions are load-bearing for the central empirical claim that O(5) truncation suffices across the adaptive loop.

minor comments (2)

[Abstract] Abstract and methods: the scaling statement 'O(Nz)' should be accompanied by a brief complexity table contrasting the Taylor approach against direct exponentiation for the largest reported matrix sizes.
[Figures] Figure captions: the molecular geometries and qubit counts for the stretched H2O case should be stated explicitly rather than referenced only in the text.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive review and the recommendation of minor revision. The single major comment is addressed point-by-point below; we will incorporate the requested clarifications and supporting data in the revised manuscript.

read point-by-point responses

Referee: [Results] Results section (fidelity and energy tables): the reported state fidelity >0.999999 and subchemical accuracy lack any description of the fidelity computation method, any error-bar analysis, and any comparison against fourth- or sixth-order truncations. These omissions are load-bearing for the central empirical claim that O(5) truncation suffices across the adaptive loop.

Authors: We agree these details should be explicit. In the revised manuscript we will add: (i) a precise description of the fidelity as |⟨ψ_exact|ψ_Taylor⟩|², with ψ_exact obtained by direct dense exponentiation (feasible for the 12–14 qubit instances); (ii) a statement that the procedure is fully deterministic, so the reported values carry no statistical error bars and are limited only by double-precision floating-point accuracy (we will report the observed numerical stability); (iii) a compact comparison of orders 4, 5, and 6 on at least one representative system (e.g., equilibrium BeH₂ under JW), showing that order 4 produces visibly larger energy deviations while order 6 yields only marginal further improvement. These additions will be placed in the Results section and will not change the central numerical claims. revision: yes

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper's core contribution is an empirical demonstration that a fixed fifth-order Taylor truncation of the unitary evolution operator suffices to maintain state fidelity >0.999999 and sub-chemical energy accuracy on the tested molecules (BeH2, H2O under JW/BK mappings). The truncation order is chosen a priori and held constant; reported fidelities and energies are obtained by direct numerical comparison against exact or high-accuracy diagonalization references that lie outside the approximation itself. No load-bearing step equates a derived quantity to a fitted parameter or self-citation by construction, and the adaptive loop results already incorporate the full operator-addition sequence. The derivation chain is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the standard Taylor series for the matrix exponential and the assumption that five terms suffice for the operator pool elements encountered in the tested molecules; no new entities or fitted parameters are introduced.

axioms (1)

domain assumption The exponential map of a skew-Hermitian operator can be approximated to sufficient accuracy by its fifth-order Taylor polynomial for the pool operators used in ADAPT-VQE.
Invoked when the paper states that the O(5) truncation retains >0.999999 fidelity.

pith-pipeline@v0.9.1-grok · 5795 in / 1312 out tokens · 26678 ms · 2026-06-30T06:35:23.325242+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

15 extracted references · 3 canonical work pages · 2 internal anchors

[1]

A variational eigenvalue solver on a photonic quantum processor,

A. Peruzzo, J. McClean, P. Shadbolt, M.-H. Yung, X.-Q. Zhou, P. J. Love, A. Aspuru-Guzik, and J. L. O’brien, “A variational eigenvalue solver on a photonic quantum processor,”Nature Communications, vol. 5, no. 1, p. 4213, 2014

2014
[2]

The theory of variational hybrid quantum-classical algorithms,

J. McClean, J. Romero, R. Babbush, and A. Aspuru-Guzik, “The theory of variational hybrid quantum-classical algorithms,”New Journal of Physics, vol. 18, p. 023023, 2016

2016
[3]

The variational quantum eigensolver: A review of methods and best practices,

J. Tilly, H. Chen, S. Cao, D. Picozzi, K. Setia, Y. Li, E. Grant, L. Wossnig, I. Rungger, G. H. Booth, and J. Tennyson, “The variational quantum eigensolver: A review of methods and best practices,”Physics Reports, vol. 986, pp. 1–128, 2022

2022
[4]

An adaptive variational algorithm for exact molecular simulations on a quantum computer,

H. R. Grimsley, S. E. Economou, E. Barnes, and N. J. Mayhall, “An adaptive variational algorithm for exact molecular simulations on a quantum computer,”Nature Communications, vol. 10, no. 1, p. 3007, 2019

2019
[5]

Über das paulischeäquivalenzverbot,

P. Jordan and E. P. Wigner, “Über das paulischeäquivalenzverbot,” Zeitschrift für Physik, vol. 47, no. 9-10, pp. 631–651, 1928

1928
[6]

Fermionic quantum computation,

S. B. Bravyi and A. Y. Kitaev, “Fermionic quantum computation,” Annals of Physics, vol. 298, no. 1, pp. 210–226, 2002

2002
[7]

Accelerating compressed sparse row matrix operations on modern heterogeneous architectures via cache- aware thread block scheduling,

L. Chen, W. Zhang, and Y. Liu, “Accelerating compressed sparse row matrix operations on modern heterogeneous architectures via cache- aware thread block scheduling,”IEEE Transactions on Parallel and Distributed Systems, vol. 36, no. 3, pp. 412–426, 2025

2025
[8]

Memory-efficient variational quan- tum simulation: Bypassing the22𝑁 density matrix barrier via adaptive chained sparse pointers,

M. Zhao, J.-W. Kim, and A. Patel, “Memory-efficient variational quan- tum simulation: Bypassing the22𝑁 density matrix barrier via adaptive chained sparse pointers,”IEEE Transactions on Quantum Engineering, vol. 7, pp. 1–12, 2026

2026
[9]

Shattering the Symmetry Trap in Fixed-Ansatz VQE: An Accelerated ADAPT-VQE Study of Three Pillar Molecules under Bravyi-Kitaev Mapping

H. K. Dipojono, “Shattering the symmetry trap in fixed-ansatz vqe: An accelerated adapt-vqe study of three pillar molecules under bravyi-kitaev mapping,”arXiv:2606.05968, 2026, (submitted to Journal of Chemical Physics)

work page internal anchor Pith review Pith/arXiv arXiv 2026
[10]

Shot-efficient adapt-vqe via reused pauli measure- ments and variance-based shot allocation,

A. Ikhtiaruddin, G. K. Sunnardianto, F. Fathurrahman, M. K. Agusta, and H. K. Dipojono, “Shot-efficient adapt-vqe via reused pauli measure- ments and variance-based shot allocation,”Physica Scripta, vol. 101, no. 25, 2026, dOI 10.1088/1402-4896/ae784b

work page doi:10.1088/1402-4896/ae784b 2026
[11]

Representation-Induced Symmetry Trapping in Adaptive Variational Quantum Simulations of Multi-Reference Topologies

H. K. Dipojono, “Representation-induced symmetry trapping in adap- tive variational quantum simulations of multi-reference topologies,” arXiv:2606.13387, 2026, (submitted to Journal of Chemical Theory and Computation)

work page internal anchor Pith review Pith/arXiv arXiv 2026
[12]

Avoiding symmetry roadblocks and minimizing the measurement overhead of adaptive variational quantum eigensolvers,

V. O. Shkolnikov, G. S. Barron, N. J. Mayhall, S. E. Economou, and E. Barnes, “Avoiding symmetry roadblocks and minimizing the measurement overhead of adaptive variational quantum eigensolvers,” Quantum, vol. 7, p. 1040, 2023

2023
[13]

Vectorized high-order taylor series appximations for sparse unitary dynamics on distributed hardware,

R. M. Williams and S. T. Davis, “Vectorized high-order taylor series appximations for sparse unitary dynamics on distributed hardware,” SIAM Journal on Scientific Computing, vol. 47, no. 2, pp. A315–A338, 2025

2025
[14]

Algorithmic mitigation of memory bandwidth choking in extreme-scale quantum circuit simulations,

H. Sato and K. Tanaka, “Algorithmic mitigation of memory bandwidth choking in extreme-scale quantum circuit simulations,”ACM Transac- tions on Computer Systems, vol. 42, no. 1, pp. 89–104, 2024. 7

2024
[15]

Gpu-accelerated large-scale sparse matrix- vector multiplication for quantum chemistry applications,

D. Filippov and M. Schmidt, “Gpu-accelerated large-scale sparse matrix- vector multiplication for quantum chemistry applications,”Journal of Computational Chemistry, vol. 45, no. 14, pp. 1120–1134, 2024

2024

[1] [1]

A variational eigenvalue solver on a photonic quantum processor,

A. Peruzzo, J. McClean, P. Shadbolt, M.-H. Yung, X.-Q. Zhou, P. J. Love, A. Aspuru-Guzik, and J. L. O’brien, “A variational eigenvalue solver on a photonic quantum processor,”Nature Communications, vol. 5, no. 1, p. 4213, 2014

2014

[2] [2]

The theory of variational hybrid quantum-classical algorithms,

J. McClean, J. Romero, R. Babbush, and A. Aspuru-Guzik, “The theory of variational hybrid quantum-classical algorithms,”New Journal of Physics, vol. 18, p. 023023, 2016

2016

[3] [3]

The variational quantum eigensolver: A review of methods and best practices,

J. Tilly, H. Chen, S. Cao, D. Picozzi, K. Setia, Y. Li, E. Grant, L. Wossnig, I. Rungger, G. H. Booth, and J. Tennyson, “The variational quantum eigensolver: A review of methods and best practices,”Physics Reports, vol. 986, pp. 1–128, 2022

2022

[4] [4]

An adaptive variational algorithm for exact molecular simulations on a quantum computer,

H. R. Grimsley, S. E. Economou, E. Barnes, and N. J. Mayhall, “An adaptive variational algorithm for exact molecular simulations on a quantum computer,”Nature Communications, vol. 10, no. 1, p. 3007, 2019

2019

[5] [5]

Über das paulischeäquivalenzverbot,

P. Jordan and E. P. Wigner, “Über das paulischeäquivalenzverbot,” Zeitschrift für Physik, vol. 47, no. 9-10, pp. 631–651, 1928

1928

[6] [6]

Fermionic quantum computation,

S. B. Bravyi and A. Y. Kitaev, “Fermionic quantum computation,” Annals of Physics, vol. 298, no. 1, pp. 210–226, 2002

2002

[7] [7]

Accelerating compressed sparse row matrix operations on modern heterogeneous architectures via cache- aware thread block scheduling,

L. Chen, W. Zhang, and Y. Liu, “Accelerating compressed sparse row matrix operations on modern heterogeneous architectures via cache- aware thread block scheduling,”IEEE Transactions on Parallel and Distributed Systems, vol. 36, no. 3, pp. 412–426, 2025

2025

[8] [8]

Memory-efficient variational quan- tum simulation: Bypassing the22𝑁 density matrix barrier via adaptive chained sparse pointers,

M. Zhao, J.-W. Kim, and A. Patel, “Memory-efficient variational quan- tum simulation: Bypassing the22𝑁 density matrix barrier via adaptive chained sparse pointers,”IEEE Transactions on Quantum Engineering, vol. 7, pp. 1–12, 2026

2026

[9] [9]

Shattering the Symmetry Trap in Fixed-Ansatz VQE: An Accelerated ADAPT-VQE Study of Three Pillar Molecules under Bravyi-Kitaev Mapping

H. K. Dipojono, “Shattering the symmetry trap in fixed-ansatz vqe: An accelerated adapt-vqe study of three pillar molecules under bravyi-kitaev mapping,”arXiv:2606.05968, 2026, (submitted to Journal of Chemical Physics)

work page internal anchor Pith review Pith/arXiv arXiv 2026

[10] [10]

Shot-efficient adapt-vqe via reused pauli measure- ments and variance-based shot allocation,

A. Ikhtiaruddin, G. K. Sunnardianto, F. Fathurrahman, M. K. Agusta, and H. K. Dipojono, “Shot-efficient adapt-vqe via reused pauli measure- ments and variance-based shot allocation,”Physica Scripta, vol. 101, no. 25, 2026, dOI 10.1088/1402-4896/ae784b

work page doi:10.1088/1402-4896/ae784b 2026

[11] [11]

Representation-Induced Symmetry Trapping in Adaptive Variational Quantum Simulations of Multi-Reference Topologies

H. K. Dipojono, “Representation-induced symmetry trapping in adap- tive variational quantum simulations of multi-reference topologies,” arXiv:2606.13387, 2026, (submitted to Journal of Chemical Theory and Computation)

work page internal anchor Pith review Pith/arXiv arXiv 2026

[12] [12]

Avoiding symmetry roadblocks and minimizing the measurement overhead of adaptive variational quantum eigensolvers,

V. O. Shkolnikov, G. S. Barron, N. J. Mayhall, S. E. Economou, and E. Barnes, “Avoiding symmetry roadblocks and minimizing the measurement overhead of adaptive variational quantum eigensolvers,” Quantum, vol. 7, p. 1040, 2023

2023

[13] [13]

Vectorized high-order taylor series appximations for sparse unitary dynamics on distributed hardware,

R. M. Williams and S. T. Davis, “Vectorized high-order taylor series appximations for sparse unitary dynamics on distributed hardware,” SIAM Journal on Scientific Computing, vol. 47, no. 2, pp. A315–A338, 2025

2025

[14] [14]

Algorithmic mitigation of memory bandwidth choking in extreme-scale quantum circuit simulations,

H. Sato and K. Tanaka, “Algorithmic mitigation of memory bandwidth choking in extreme-scale quantum circuit simulations,”ACM Transac- tions on Computer Systems, vol. 42, no. 1, pp. 89–104, 2024. 7

2024

[15] [15]

Gpu-accelerated large-scale sparse matrix- vector multiplication for quantum chemistry applications,

D. Filippov and M. Schmidt, “Gpu-accelerated large-scale sparse matrix- vector multiplication for quantum chemistry applications,”Journal of Computational Chemistry, vol. 45, no. 14, pp. 1120–1134, 2024

2024